Deep Learning Approach for Medicinal Plant Identification and Therapeutic Use Prediction

The therapeutic nature of medicinal plants and their ability to heal many diseases raises the need for their automatic identification. However, the preservation and transmission of traditional medicinal knowledge encounter substantial problems due to factors such as professional secrecy, wrong cultural beliefs and lack of experts on the field. Consequently, there is a pressing risk of traditional medicinal knowledge fading away, resulting in a decline in expertise and an increased likelihood of errors in a field that demands serious attention and care. This work explores deep learning and proposes a deep learning approach for medicinal plant identification and to predict therapeutic use of medicinal plants from their leaf image. The transfer learning approach along with the pre-trained neural networks such as VGG16, InceptionV3 and Xception architectures were employed to extract features from the input leaf images. The Softmax and support vector machine(SVM) were used to classify the features generated by the three feature extractor model. The InceptionV3 architecture outperformed the state-of-the-art pre-trained models by achieving 99.17% average validation accuracy of Support Vector Machine (SVM) and softMax classifier) and the maximum validation accuracy of 99.34 is obtained using SVM. The average validation accuracy of the two classifier on each feature extractor model is used to select the better feature extractor model. To improve performance the SVM classifier was tuned using the Bayesian optimization technique and the proposed model InceptionV3 was validated with our custom datasets and achieved 99.34% prediction accuracy. The finding of this work showcases transfer learning-based InceptionV3 feature with support vector machine classifier(SVM), can effectively predict the therapeutic use of medicinal plants from leaf images with high accuracy and offers a promising solution to the fading traditional medicinal knowledge.